maedhroz commented on code in PR #3916:
URL: https://github.com/apache/cassandra/pull/3916#discussion_r1964524913
##########
src/java/org/apache/cassandra/service/reads/ReplicaFilteringProtection.java:
##########
@@ -538,14 +538,19 @@ private UnfilteredPartitionIterator fetchFromSource()
Tracing.trace("Requesting {} rows in partition {} from {} for
replica filtering protection",
clusterings.size(), key, source);
- // build the read command taking into account that we could be
requesting only in the static row
- DataLimits limits = clusterings.isEmpty() ?
DataLimits.cqlLimits(1) : DataLimits.NONE;
- ClusteringIndexFilter filter = unresolvedStatic ?
command.clusteringIndexFilter(key) : new
ClusteringIndexNamesFilter(clusterings, command.isReversed());
+ // If there is an unresolved static column, we must fetch the
entire partition, as static column predicates
+ // may produce row matches across the entire partition. If there
are only non-static rows to complete, we
+ // query the partition specifically for the corresponding
cluterings by name. In either case, we do not
+ // provide a limit. (In the unresolved static case, we have no way
of knowing how many stale rows we might
+ // read on a silent replica before finding a live one.)
+ ClusteringIndexFilter filter = unresolvedStatic ?
command.clusteringIndexFilter(key)
+ : new
ClusteringIndexNamesFilter(clusterings, command.isReversed());
Review Comment:
This is the crux of the patch. I went back and forth a bit, but I don't know
if we can avoid fetching the rest of the partition (in the current data range).
In practice, we're not going to be using ridiculously small fetch sizes, so I
don't know how much of a problem this will actually be, but the viral nature of
static columns is what it is.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
